Nadia Makarevich explores the capabilities of AI in debugging a React/Next.js app, using a hands-on approach to test its effectiveness against real-world issues. She sets up a project on GitHub, which includes various components and API endpoints, but intentionally introduces bugs, particularly focusing on a broken User Profile page. The goal is to see how well a large language model (LLM) can identify and resolve these issues compared to a manual debugging process. After encountering a "Something went wrong" error on the user page, she inputs the error details into the LLM. The AI identifies the problem: missing fields in the user object that violate the validation schema defined by Zod. The model suggests adding the missing fields, which initially resolves the issue. Makarevich then conducts her own investigation to validate the LLM's findings. She confirms that the error stems from the API returning incomplete data, matching the LLM's diagnosis. The article emphasizes the importance of understanding both AI capabilities and the underlying code. While the LLM managed to fix the immediate issue, Makarevich highlights that there are two potential solutions: ensuring that the API always returns the required fields or modifying the schema to make those fields optional. The choice between these solutions hinges on the broader context of the application and its data integrity requirements.